What can wait together?
Group small tasks. Keep every original.
Ready. The first run downloads the open model.
Inference runs in your browser. Initial model and library downloads contact Hugging Face and jsDelivr; task text is not sent to an inference server. English tasks only. Up to 30 tasks.
How suggestions work
MiniLM compares meanings. Tasks only merge if every pair is similar, their context and selected day match, and their combined estimate is at most 30 minutes. “Separate” always wins. A limited English word check flags possible restrictions and reversals; it cannot understand every contradiction. Review all suggestions.
No deadlines are inferred. No task is deleted or completed. Minutes are your estimates, not measured savings. Editing inputs invalidates previous suggestions.
Open model provenance